Pioneering Advanced Deep Multi-Task Learning Model for Precise COVID-19 Diagnosis

To address the urgent need for intelligent systems capable of simultaneously identifying COVID-19 cases and delineating lung infection regions in CT scan images, we introduce a multi-task architecture. Initially, our model, an enhanced U-Net with a classification branch, aims to balance segmentation and classification tasks using binary preprocessing algorithms. The CBAM attention module further improves task coherence. Rigorous evaluation demonstrates our model's superiority over previous studies, achieving an impressive 95% classification accuracy and a Dice index of 88.37 ± 0.01 for segmentation, supported by robust metrics. We also extend our multi-task framework to complex U-Net++ and ResUNet architectures with positive outcomes. Future exploration includes integrating ensemble learning within this framework. Our work signifies significant progress in enhancing diagnostic capabilities in complex medical image analysis, elevating standards for COVID-19 detection and lung infection analysis through multi-task learning and shared feature extraction.